Exploiting Timelines to Enhance Multi-document Summarization

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 923–933, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Exploiting Timelines to Enhance Multi-document Summarization Jun-Ping Ng1,2, Yan Chen3, Min-Yen Kan2,4, Zhoujun Li3 1DSO National Laboratories, Singapore 2School of Computing, National University of Singapore, Singapore 3State Key Laboratory of Software Development Environment, Beihang University, China 4Interactive and Digital Media Institute, National University of Singapore, Singapore njunping@dso.org.sg Abstract We study the use of temporal information in the form of timelines to enhance multi- document summarization. We employ a fully automated temporal processing sys- tem to generate a timeline for each in- put document. We derive three features from these timelines, and show that their use in supervised summarization lead to a significant 4.1% improvement in ROUGE performance over a state-of-the-art base- line. In addition, we propose TIMEMMR, a modification to Maximal Marginal Rel- evance that promotes temporal diversity by way of computing time span similar- ity, and show its utility in summarizing certain document sets. We also propose a filtering metric to discard noisy timelines generated by our automatic processes, to purify the timeline input for summariza- tion. By selectively using timelines guided by filtering, overall summarization perfor- mance is increased by a significant 5.9%. 1 Introduction There has been a good amount of research in- vested into improving the temporal interpretation of text. Besides the increasing availability of an- notation standards (e.g., TIMEML (Pustejovsky et al., 2003a)) and corpora (e.g., TIDES (Ferro et al., 2000), TimeBank (Pustejovsky et al., 2003b)), the community has also organized three success- ful evaluation workshops — TempEval-1 (Verha- gen et al., 2009), -2 (Verhagen et al., 2010), and -3 (Uzzaman et al., 2013). As the state-of-the- art improves, these workshops have moved away from the piecemeal evaluation of individual tem- poral processing tasks and towards the evaluation of complete end-to-end systems in TempEval-3. We believe our understanding of the temporal in- formation found in text is sufficiently robust, and that there is an opportunity to now leverage this in- formation in downstream applications. In this pa- per, we present our work in incorporating the use of such temporal information into multi-document summarization. The goal of multi-document summarization is to generate a summary which includes the main points from an input collection of documents with minimal repetition of similar points. We hope to improve the quality of the summaries that are gen- erated by considering temporal information found in the input text. To motivate how temporal in- formation can be useful in summarization, let us refer to Figure 1. The three sentences describe a recent cyclone and a previous one which happened in 1991. Recognizing that sentence (3) is about a storm that had happened in the past is important when writing a summary about the recent storm, as it is not relevant and can likely be excluded. It is reasonable to expect that a collection of documents about the recent storm will contain more references to it, compared with …

Content truncated to 500 words